[ARM] Fix invalid VLDM/VSTM access when targeting Big Endian with NEON

The instructions VLDM/VSTM can only access word-aligned memory
locations and produce alignment fault if the condition is not met.

The compiler currently generates VLDM/VSTM for v2f64 load/store
regardless the alignment of the memory access. Instead, if a v2f64
load/store is not word-aligned, the compiler should generate
VLD1/VST1. For each non double-word-aligned VLD1/VST1, a VREV
instruction should be generated when targeting Big Endian.

Differential Revision: https://reviews.llvm.org/D25281

llvm-svn: 283763

1 files changed, 12 insertions, 2 deletions

diff --git a/llvm/lib/Target/ARM/ARMInstrNEON.td b/llvm/lib/Target/ARM/ARMInstrNEON.td
index defef4ea907..b5fa8e999e2 100644
--- a/llvm/lib/Target/ARM/ARMInstrNEON.td
+++ b/llvm/lib/Target/ARM/ARMInstrNEON.td
@@ -610,14 +610,14 @@ def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
 def VLDMQIA
   : PseudoVFPLdStM<(outs DPair:$dst), (ins GPR:$Rn),
                     IIC_fpLoad_m, "",
-                   [(set DPair:$dst, (v2f64 (load GPR:$Rn)))]>;
+                   [(set DPair:$dst, (v2f64 (word_alignedload GPR:$Rn)))]>;
 
 // Use VSTM to store a Q register as a D register pair.
 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
 def VSTMQIA
   : PseudoVFPLdStM<(outs), (ins DPair:$src, GPR:$Rn),
                     IIC_fpStore_m, "",
-                   [(store (v2f64 DPair:$src), GPR:$Rn)]>;
+                   [(word_alignedstore (v2f64 DPair:$src), GPR:$Rn)]>;
 
 // Classes for VLD* pseudo-instructions with multi-register operands.
 // These are expanded to real instructions after register allocation.
@@ -6849,6 +6849,16 @@ let Predicates = [IsBE] in {
   def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (VREV64q32 QPR:$src)>;
 }
 
+// Use VLD1/VST1 + VREV for non-word-aligned v2f64 load/store on Big Endian
+def : Pat<(v2f64 (byte_alignedload addrmode6:$addr)),
+          (VREV64q8 (VLD1q8 addrmode6:$addr))>, Requires<[IsBE]>;
+def : Pat<(byte_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
+          (VST1q8 addrmode6:$addr, (VREV64q8 QPR:$value))>, Requires<[IsBE]>;
+def : Pat<(v2f64 (hword_alignedload addrmode6:$addr)),
+          (VREV64q16 (VLD1q16 addrmode6:$addr))>, Requires<[IsBE]>;
+def : Pat<(hword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
+          (VST1q16 addrmode6:$addr, (VREV64q16 QPR:$value))>, Requires<[IsBE]>;
+
 // Fold extracting an element out of a v2i32 into a vfp register.
 def : Pat<(f32 (bitconvert (i32 (extractelt (v2i32 DPR:$src), imm:$lane)))),
           (f32 (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane)))>;